Technology

AI experts demand independent oversight and protections for third-party safety evaluators

A public letter signed by over 100 researchers calls for clear standards and whistleblower protections.

Updated Sep 18, 2026 · 19:44 UTC2 reports

The short version

A coalition of more than 100 AI researchers and evaluators released a public letter demanding guaranteed independence, transparency, and whistleblower protections for third-party evaluators testing advanced models.[Business Insider · CNBC]
The letter follows a proposal from Anthropic CEO Dario Amodei to grant select external evaluators employee-like access to internal systems and unreleased technology.[CNBC]
Tech leaders like Sam Altman and Satya Nadella supported the access proposal, but evaluators say clear standards are needed so companies cannot suppress findings or retaliate.[CNBC]

Over 100 AI experts, including Geoffrey Hinton and Stuart Russell, signed a public letter organized by the AI Evaluator Forum.[Business Insider · CNBC]
The letter demands that embedded safety evaluators have scientific objectivity, independence, transparency, and protections against company interference.[Business Insider · CNBC]
Anthropic CEO Dario Amodei proposed granting third-party evaluators employee-like access to company systems, internal data, and unreleased models.[CNBC]
Industry figures including Sam Altman, Satya Nadella, and Elon Musk expressed support for the concept of embedded evaluators.[CNBC]

AI companies have not detailed the operational logistics of the evaluator program, including which groups will be selected or how deeply they can inspect internal systems.[CNBC]
It remains uncertain whether foundation model developers will adopt the standards proposed in the public letter.[CNBC]

Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.